kern/173541: load average 0.60 at 100% idle

[Johan Broman]

The following reply was made to PR kern/173541; it has been noted by GNATS.

From: Johan Broman <je.broman at gmail.com>
To: bug-followup at freebsd.org
Cc:  
Subject: Re: kern/173541: load average 0.60 at 100% idle
Date: Fri, 4 Jan 2013 00:06:16 +0100

 --047d7b5d95370023f704d26a691e
 Content-Type: text/plain; charset=ISO-8859-1
 
 Hi!
 
 I see. I got around the problem in FreeBSD 9.1 by changing the clock
 source. Like this:
 
 # sysctl -w kern.eventtimer.timer=LAPIC
 
 The load avg dropped to 0.00 again. You can try the different clock sources
 available to you and see what works best. Some sources might increase the
 number of interrupts (like the RTC) and some might cause more context
 switching or CPU load. For me the LAPIC works best. You can list your clock
 sources using:
 
 # sysctl kern.eventtimer.choice
 
 From what I understand, measuring time can be tricky because new systems
 can regulate core frequency on the fly (in HW) and virtualization also
 increases the difficulty...
 
 It seems the order and/or weight of the clock sources has changed. When I
 have a chance I will reboot into the old kernel and see what has been
 changed. I'm totally new to FreeBSD so I don't know the normal handling of
 these bugs but I'll put the author of the patch on cc as well :)
 
 Good luck!
 Johan
 
 --047d7b5d95370023f704d26a691e
 Content-Type: text/html; charset=ISO-8859-1
 Content-Transfer-Encoding: quoted-printable
 
 Hi!<br><br>I see. I got around the problem in FreeBSD 9.1 by changing the c=
 lock source. Like this:<br><br># sysctl -w kern.eventtimer.timer=3DLAPIC<br=
 ><br>The load avg dropped to 0.00 again. You can try the different clock so=
 urces available to you and see what works best. Some sources might increase=
  the number of interrupts (like the RTC) and some might cause more context =
 switching or CPU load. For me the LAPIC works best. You can list your clock=
  sources using:<br>
 <br># sysctl kern.eventtimer.choice<br><br>From what I understand, measurin=
 g time can be tricky because new systems can regulate core frequency on the=
  fly (in HW) and virtualization also increases the difficulty...<br><br>
 It seems the order and/or weight of the clock sources has changed. When I h=
 ave a chance I will reboot into the old kernel and see what has been change=
 d. I'm totally new to FreeBSD so I don't know the normal handling o=
 f these bugs but I'll put the author of the patch on cc as well :)<br>
 <br>Good luck!<br>Johan<br><br>
 
 --047d7b5d95370023f704d26a691e--